A Multi-Head Relevance Weighting Framework For Learning Raw Waveform Audio Representations

TL;DR

This paper introduces a multi-head relevance weighting framework that enhances raw waveform audio representations by applying learnable filterbanks and relevance masks, leading to significant improvements in audio classification tasks.

Contribution

The novel multi-head relevance weighting approach improves raw waveform audio representations by integrating learnable filterbanks with relevance masks for better classification performance.

Findings

Achieved 10% and 23% relative improvements over mel-spectrogram baseline on DCASE2020 and USC datasets.

Relevance weights provide insights into learned representations.

Framework effectively enhances raw waveform audio classification.

Abstract

In this work, we propose a multi-head relevance weighting framework to learn audio representations from raw waveforms. The audio waveform, split into windows of short duration, are processed with a 1-D convolutional layer of cosine modulated Gaussian filters acting as a learnable filterbank. The key novelty of the proposed framework is the introduction of multi-head relevance on the learnt filterbank representations. Each head of the relevance network is modelled as a separate sub-network. These heads perform representation enhancement by generating weight masks for different parts of the time-frequency representation learnt by the parametric acoustic filterbank layer. The relevance weighted representations are fed to a neural classifier and the whole system is trained jointly for the audio classification objective. Experiments are performed on the DCASE2020 Task 1A challenge as well as the Urban Sound Classification (USC) tasks. In these experiments, the proposed approach yields relative improvements of 10% and 23% respectively for the DCASE2020 and USC datasets over the mel-spectrogram baseline. Also, the analysis of multi-head relevance weights provides insights on the learned representations.